SynpickVP Dataset

from SynpickVP import SynpickVP
import visualizations as visualizations

import torch
from matplotlib import pyplot as plt

%load_ext autoreload
%autoreload 2

Instanciating Dataset Splits

We instanciate the training, validation and test datasets.

Keep in mind that validation and test always have num_frames = 50 for consistency and reproducibility

train_db = SynpickVP(
        split="train",
        num_frames=10,
        seq_step=2,
        img_size=(136, 240),
    )

val_db = SynpickVP(
        split="val",
        num_frames=10,
        seq_step=2,
        img_size=(136, 240),
    )

test_db = SynpickVP(
        split="test",
        num_frames=10,
        seq_step=2,
        img_size=(136, 240),
    )

100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████| 545957/545957 [00:04<00:00, 123332.49it/s]

SynpickVP test-sequences have 50 frames. Your num_frames = 10 will be overridden

100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████| 31293/31293 [00:00<00:00, 41141.71it/s]

SynpickVP test-sequences have 50 frames. Your num_frames = 10 will be overridden

100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████| 409568/409568 [00:10<00:00, 38877.34it/s]

num_seqs = len(train_db)
print(f"SynpickVP training set with seq_len={train_db.num_frames} has {num_seqs} sequences")

num_seqs = len(val_db)
print(f"SynpickVP validation set with seq_len={val_db.num_frames} has {num_seqs} sequences")

num_seqs = len(test_db)
print(f"SynpickVP test set with seq_len={test_db.num_frames} has {num_seqs} sequences")

SynpickVP training set with seq_len=10 has 42101 sequences
SynpickVP validation set with seq_len=50 has 121 sequences
SynpickVP test set with seq_len=50 has 1408 sequences

Visualizations

First Train Sequence

cur_data = train_db[0]
frames = cur_data["frames"]
maps = cur_data["segmentation"]
instances = cur_data["instance"]

(10, 135, 240)
(10, 135, 240)

_ = visualizations.visualize_sequence(frames, n_cols=5)

disp = visualizations.overlay_segmentations(
        frames=frames,
        segmentations=maps,
        colors=visualizations.COLORS,
        num_classes=train_db.NUM_CLASSES,
        alpha=0.5
    )
_ = visualizations.visualize_sequence(disp, n_cols=5)

disp = visualizations.overlay_instances(
        frames=frames,
        instances=instances,
        colors=visualizations.COLORS,
        alpha=0.5
    )
_ = visualizations.visualize_sequence(disp, n_cols=5)

First Test Sequence

cur_data = test_db[2]
frames = cur_data["frames"][:20]
maps = cur_data["segmentation"][:20]
instances = cur_data["instance"][:20]

_ = visualizations.visualize_sequence(frames, n_cols=10)

disp = visualizations.overlay_segmentations(
        frames=frames,
        segmentations=maps,
        colors=visualizations.COLORS,
        num_classes=train_db.NUM_CLASSES,
        alpha=0.5
    )
_ = visualizations.visualize_sequence(disp, n_cols=10)

disp = visualizations.overlay_instances(
        frames=frames,
        instances=instances,
        colors=visualizations.COLORS,
        alpha=0.7
    )
_ = visualizations.visualize_sequence(disp, n_cols=10)

Angel Villar-Corrales

• Email: villar@ais.uni-bonn.de
• Website: angelvillarcorrales.com